The challenge of simulating large materials has been tackled with reinforcement learning prediction models, as traditional simulation methods are only able to simulate small and simple molecules. This approach allows for better approximations of electron cloud interactions in specific shapes, as described in a Nature paper.